Keir Fraser [Wed, 15 Dec 2010 11:56:25 +0000 (11:56 +0000)]
x86: x2apic pre-enabled but intr-remapping is not enabled
Make it aligned with Linux kernel.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Removed unnecessary bits from the original patch, and removed
intremap_enabled() with its only caller gone.
Keir Fraser [Wed, 15 Dec 2010 11:55:48 +0000 (11:55 +0000)]
x86: increase MAX_LOCAL_APIC
otherwise apicid_to_node[MAX_LOCAL_APIC] will be overrun if apicid >
255. After patch, the mapping get right.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Make this and also MAX_MADT_ENTRIES loosely depend on NR_CPUS. Tie
MAX_APICS to MAX_LOCAL_APIC. Fix initializer of x86_acpiid_to_apicid[]
to match the array member type of u32, as well as all checks in
readers of this array and x86_cpu_to_apicid[].
While the adjustment to xen_vcpu_physid_to_x86_{acpi,apic}id() is not
backward compatible, I think it should still be done this way as the
former reserving of values beyond 0xff should never have been part of
the interface. If considered impossible, a second best solution would
appear to be to make the macros depend on __XEN_INTERFACE_VERSION__.
Keir Fraser [Wed, 15 Dec 2010 11:52:14 +0000 (11:52 +0000)]
x86: mpparse and cstate need to use 32bit apic id
Instead of going with mpc_config_processor struct.
that field ony have 8 bits.
We should not change that struct, because it is shared with mptable.
Also need to increase MAX_APICS.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Rather than using a fixed value of 512, make this scale with NR_CPUS
(which obviously still doesn't cover all theoretically possible
systems, but at least allows some build time control).
Keir Fraser [Wed, 15 Dec 2010 11:01:59 +0000 (11:01 +0000)]
x86 hvm: Emulate MSR_IA32_TSC_DEADLINE
Accesses to MSR_IA32_TSC_DEADLINE are trapped, with value stored in a
new field vlapic->hw.tdt_msr. vlapic->pt is reused in one shot mode
for vtdt to trigger expire events.
For details, please refer to the Intel Architectures Software
Developer's Manual 3A, 10.5.4.1 TSC-Deadline Mode.
Keir Fraser [Wed, 15 Dec 2010 10:47:05 +0000 (10:47 +0000)]
ept: Remove lock in ept_get_entry, replace with access-once semantics.
This mirrors the RVI/shadow situation, where p2m read access is
lockless because it's done in the hardware (linear map of the p2m
table).
This fixes the original bug (call it bug A) without introducing bug B
(a deadlock).
Bug A was caused by a race when updating p2m entries: between testing
if it's valid, and testing if it's populate-on-demand, it may have
been changed from populate-on-demand to valid.
My original patch simply introduced a lock into ept_get_entry, but
that caused bug B, caused by circular locking order: p2m_change_type
[grabs p2m lock] -> set_p2m_entry -> ept_set_entry ->
ept_set_middle_level -> p2m_alloc [grabs hap lock] write cr4 ->
hap_update_paging_modes [grabes hap lock] -> hap_update_cr3 ->
gfn_to_mfn -> ept_get_entry -> [grabs p2m lock]
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Keir Fraser [Wed, 15 Dec 2010 10:27:18 +0000 (10:27 +0000)]
tmem: two wrongs (or three lefts and a wrong) make a right
These two bugs apparently complement each other enough that
they escaped problems in my testing, but eventually gum
up the works and are obviously horribly wrong.
Found while developing tmem for native Linux.
Signed-off-by: Dan Magenheimer <dan.magenheimer@oracle.com>
Keir Fraser [Wed, 15 Dec 2010 10:21:05 +0000 (10:21 +0000)]
hvm save: Introduce hvm_load_entry_zeroextend().
In certain cases this will allow us to load old HVM save images where
an HVM saved chunk has subsequently been extended with new
fields. Rather than fail to load the chunk, we can pad the extended
structure with zeroes, if the caller knows how to handle that.
Signed-off-by: Keir Fraser <keir@xen.org> Acked-by: Tim Deegan <Tim.Deegan@citrix.com>
libxl: do not assume target and freemem-slack are written at the same time
This improves robustness in some pathological configurations.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com> Tested-by: Christoph Egger <Christoph.Egger@amd.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Introduce libxl_need_xenpv_qemu to detect if the caller needs to create
a pv qemu instance (using libxl_create_xenpv_qemu).
A positive reply depends on the number of pv console and vfbs, and the
type of disk backends.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Olaf Hering [Tue, 14 Dec 2010 18:50:49 +0000 (18:50 +0000)]
xenpaging: do not use DPRINTF/ERROR if xch handle is unavailable
Fix DPRINTF/ERROR usage. Both macros reference a xch variable in local scope.
If xc_interface_open fails and after xc_interface_close, both can not be used
anymore. Use standard fprintf for this case.
Remove the code to print the exit value, its not really useful.
Its a left-over for debugging from an earlier patch.
Signed-off-by: Olaf Hering <olaf@aepfle.de> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Olaf Hering [Tue, 14 Dec 2010 18:48:44 +0000 (18:48 +0000)]
xenpaging: print DPRINTF ouput if XENPAGING_DEBUG is in environment
No DPRINTF output is logged because the default loglevel is to low in
libxc. Recognize the XENPAGING_DEBUG environment variable to change the
default loglevel at runtime.
Signed-off-by: Olaf Hering <olaf@aepfle.de> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Ian Jackson [Tue, 14 Dec 2010 17:38:18 +0000 (17:38 +0000)]
tools/hotplug: Do not mind if "ip link set" fails on vif backends
Most versions of netback do not support setting the MAC address. This
means that c/s 937488219719 causes the hotplug script to break for
vifs, even as it fixes the bridge-wrong-MAC-address problem for tap
devices (used for emulated nics).
The mac-setting operation is not necessary for vifs since they are
hardcoded to fe:ff* anyway.
As a band-aid, add "|| true" to the call to ip link, so that this
error does not cause the hotplug operation to fail. There will still
be an error message printed until we can fix this better.
Suggested-by: Ian Campbell <Ian.Campbell@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Kouya Shimura [Tue, 14 Dec 2010 17:15:48 +0000 (17:15 +0000)]
tools/hotplug/Linux: forced to release lock if holder process is gone
When a script using locking.sh is stopped by ctrl-C, the lock file remains.
We have to wait 100 seconds for releasing the lock at the next time.
Signed-off-by: Kouya Shimura <kouya@jp.fujitsu.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Ian Jackson [Tue, 14 Dec 2010 16:56:54 +0000 (16:56 +0000)]
xenstore: new XS_OPEN_SOCKETONLY flag; honour "-s" on tools' cmdline
The "-s" option to xenstore-ls is used by the xencommons startup
script to check whether xenstored is already running, before starting
it.
Since 22498:a40455ae9ad3, "-s" has been a no-op, and libxenstore will
always fall back to using xenbus. The combined effect is that the
xencommons script deadlocks: xenstore-ls hangs waiting for xenstored,
which isn't started by xencommons because xencommons is waiting for
xenstore-ls.
In this patch, we:
* Introduce a new XS_OPEN_SOCKETONLY flag which disables the
fallback behaviour;
* Make the xenstore command line tools use the new xs_open call
rather than the old, deprecated xs_open_* calls (which are
now identical).
* Plumb the xenstore command line tools "-s" option to set the
XS_OPEN_SOCKETONLY flag.
* Change the type of the XS_OPEN_* flags so that they naturally have
type unsigned long.
The "-s" option to xenstore-ls et al, and the XS_OPEN_SOCKETONLY flag,
are intended for use by toolstack infrastructure and should not
normally be used by higher-level code.
Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Tue, 14 Dec 2010 09:54:10 +0000 (09:54 +0000)]
x86/iommu: account for necessary allocations when calculating Dom0's initial allocation size
As of c/s 21812:e382656e4dcc, IOMMU related allocations for Dom0
happen only after it got all of its memory allocated, and hence the
reserve (mainly for setting up its swiotlb) may get exhausted without
accounting for the necessary allocations up front.
While not precise, the estimate has been found to be within a couple
of pages for the systems it got tested on.
For the calculation to be reasonably correct, this depends on the
patch titled "x86/iommu: don't map RAM holes above 4G" sent out
yesterday.
Keir Fraser [Tue, 14 Dec 2010 09:52:57 +0000 (09:52 +0000)]
x86/iommu: don't map RAM holes above 4G
Matching the comment in iommu_set_dom0_mapping(), map only actual RAM
from the address range starting at 4G. It's not clear though whether
that comment is actually correct (which is why I'm sending this as
RFC), but it is certain that on systems with sparse physical memory
map we're currently wasting a potentially significant amount of memory
for setting up IOMMU page tables that will never be used.
The main question is what happens for MMIO ranges living above 4G. Of
course, the same issue would currently exist for any such ranges
sitting beyon the end of RAM.
Anthony PERARD [Mon, 13 Dec 2010 17:59:45 +0000 (17:59 +0000)]
libxl: Makes libxl be able to call Qemu upstream for XenPV guest.
In libxl_build_device_model_args_new:
- Adds -xen-attach options to the list of arguments to Qemu.
- Adds -vga xenfb options when vnc and sdl are not set.
- Remove disk list from the command line for XenPV as they will be
read from xenstore by Qemu.
Anthony PERARD [Mon, 13 Dec 2010 17:59:02 +0000 (17:59 +0000)]
libxl: strdup disk path before put it in qemu args array.
In libxl_build_device_model_args_new, the path to the disk image are
freeed before there was actually use to make the arguments list of Qemu.
The patch strdups it.
This patch also changes argv[0] of the device model.
Now, it is the conventional argv[0], so is value come from
info->device_model.
Chunyan Liu [Mon, 13 Dec 2010 17:39:27 +0000 (17:39 +0000)]
Fix /vm/uuid xenstore leak on tapdisk2 device cleanup
While doing block-detach blktap2 disk, tap2 device info is not cleared from
xenstore /vm/uuid/xxx. The reason is in xen-hotplug-cleanup script: when $vm_dev
does not exist, $(xenstore-read "$vm_dev" 2>/dev/null) is also "", won't enter
the block. So, change to use cmd return value to check existence.
Doesn't actually appear to be used anywhere but is defined for other
OSes.
The NetBSD evtchn.h contains comments "Return allocated port" for
several ioctls which currently return the allocated port as a member
of the argument structure and not as the ioctl return value (I think
this is a cut and paste error).
Signed-off-by: Ian Campbell <ian.campbell@citrix.com> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com> Acked-by: Christoph Egger <Christoph.Egger@amd.com>
committer: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Keir Fraser [Fri, 10 Dec 2010 19:38:30 +0000 (19:38 +0000)]
etherboot: Download ipxe from XEN_EXTFILES_URL
Allows us to build even if the ipxe git server is down. We still fall
back to the git server if we cannot download a suitably named tarball
from our own URL.
Ian Campbell [Fri, 10 Dec 2010 18:44:07 +0000 (18:44 +0000)]
tools/hotplug/Linux: Ensure tap devices receive a dummy MAC address.
If a tap device is not given an explicit MAC address it will generate
one randomly.
The behaviour of the Linux bridge is to pickup the lowest MAC address
of any port for use in for ARP, STP etc. If the tap device's randomly
generated MAC address happens to be the lowest then this can cause all
manner of strange networking glitches in both domain 0 and guests when
the bridge suddenly takes over from the previously used MAC address.
We choose FE:FF:FF:FF:FF:FF as it the numerically largest
non-broadcast address. This ensures that the physical NIC device's
port will have the lowest MAC address and therefore be the one picked
up by the bridge.
vif devices already have a MAC address of FE:FF:FF:FF:FF:FF set by
netback already but there is no harm in forcing it a second time in
the hotplug script.
tap devices are added by the "add" event and therefore we should call
setup_bridge_port then as well as for "online" which is caused by vif
devices.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Ian Campbell [Fri, 10 Dec 2010 18:30:45 +0000 (18:30 +0000)]
libxl: "Device model is a stubdom" is not an error
Signed-off-by: Ian Campbell <ian.campbell@citrix.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Ian Campbell [Fri, 10 Dec 2010 18:29:50 +0000 (18:29 +0000)]
libxl: return an error if connection to xenstore fails
Signed-off-by: Ian Campbell <ian.campbell@citrix.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Jan Beulich [Fri, 10 Dec 2010 18:27:17 +0000 (18:27 +0000)]
libxc: correct bounce direction for debug-key handling
Signed-off-by: Jan Beulich <jbeulich@novell.com> Acked-by: Ian Campbell <Ian.Campbell@citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Jan Beulich [Fri, 10 Dec 2010 18:08:19 +0000 (18:08 +0000)]
xend: fix "xm block-detach 0 ..." for extended-ID devices
Simply taking stat()'s st_rdev doesn't work here, as the minor is
split into two parts, the major is present, and the "extended" bit
isn't set.
Rather than fixing this in a way that would likely be OS-dependent,
simply remove the access to the device file, and instead just parse
the provided string (as is done e.g. for block-attach).
Signed-off-by: Jan Beulich <jbeulich@novell.com> Acked-by: Ian Jackson <ian.jackson@eu.citrix.com> Signed-off-by: Ian Jackson <ian.jackson@eu.citrix.com>
Keir Fraser [Fri, 10 Dec 2010 11:32:19 +0000 (11:32 +0000)]
x86 acpi: Follow Windows behaviour more closely during reset.
This follows some changes proposed for upstream Linux:
1. Do not check the FADT reset register size/offset
2. Try ACPI poking twice during our reset attempt sequence
Hopefully this will help us reset reliably on a wider range of
platforms.
Keir Fraser [Fri, 10 Dec 2010 10:51:04 +0000 (10:51 +0000)]
credit2: Don't migrate cpus unnecessarily
Modern processors have one or two L3's per socket, and generally only
one core per L2. Credit2's design relies on having credit shared
across several. So as a first step to sharing a queue across L3 while
avoiding excessive cross-L2 migration
Step one: If the vcpu's current cpu is acceptable, just run it there.
Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>
Tim Deegan [Wed, 8 Dec 2010 10:46:31 +0000 (10:46 +0000)]
x86/mm: change ASSERTs to BUG_ONs in mem_sharing.c
These two ASSERTs have important side-effects so make them into BUG_ONs
consistent with the rest of the file.
Bug found by Jui-Hao Chiang <juihaochiang@gmail.com>.
Keir Fraser [Tue, 7 Dec 2010 18:32:04 +0000 (18:32 +0000)]
x86: remove BUG_ON() from QUIRK_IOAPIC_*_REGSEL handler
Since (non-pvops, 32-bit only up to 2.6.27) Linux would report "BAD"
unconditionally on all SiS chipset versions (it only looks for a PCI
device at 0000:00:00.0 with SiS as the vendor), we must not crash if
the report on a 64-bit hypervisor doesn't match the #define (which is
zero).
While we could honor the quirk indication even on 64-bit, it doesn't
seem worthwhile, as there's no evidence that newer SiS chipsets
(supporting 64-bit CPUs) are actually affected.
This should also address bug 1687 (mis-reported, however, afaict).
Keir Fraser [Tue, 7 Dec 2010 18:26:38 +0000 (18:26 +0000)]
amd xsave: Move xsave initialization code to a common place
This patch moves xsave/xrstor code to CPU common file. First of all,
it prepares xsave/xrstor support for AMD CPUs. Secondly, Xen would
crash on __context_switch() without this patch on xsave-capable AMD
CPUs. The crash was due to cpu_has_xsave reports true in domain.c
while xsave space wasn't initialized.
Keir Fraser [Tue, 7 Dec 2010 18:24:12 +0000 (18:24 +0000)]
x86 hvm: x2APIC emulation
This patch would enable Xen to handle x2APIC MSR accessing of HVM
guest, which is faster(avoid decoding of MMIO accessing). The credit
comes to Gleb Natapov who complete the work for KVM.
Have tested with 4 vcpus guest, with/without x2apic support.
From: Sheng Yang <sheng.yang@intel.com> Signed-off-by: Keir Fraser <keir@xen.org>
Keir Fraser [Fri, 3 Dec 2010 06:37:48 +0000 (06:37 +0000)]
minios: reverse layering of xc vs minios fd close
Having minios close() call back into the libxc core close routines is
backwards and unexpected. On every other OS the libxc core close
routine calls close().
Export minios specific functions from the minios libxc code to
implement fd closing for each type of xc file handle and simply call
close() in the core close routine.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Keir Fraser [Wed, 1 Dec 2010 20:10:27 +0000 (20:10 +0000)]
x86: fix IRQ migration when using directed EOI (broken with c/s 20465)
In directed-EOI mode, there is no chance to do the migration in
mask_and_ack_level_ioapic_irq(), as the remote IRR bit can't possibly
be clear after issuing the EOI to the LAPIC. Consequently, there's no
point to even try. Instead, migration must be done in
end_level_ioapic_irq(), and it requires masking the interrupt source
prior to issuing the EOI to the IO-APIC.
Keir Fraser [Tue, 30 Nov 2010 11:34:08 +0000 (11:34 +0000)]
x86 hvm: Do not overwrite boot-cpu capability data on VMX/SVM startup.
Apparently required back in the earliest days of Xen, we now properly
initialise CPU capabilities early during bootstrap. Re-writing
capability data later now causes problems if specific features have
been deliberately masked out.
Thanks to Weidong Han at Intel for finding such a bug where XSAVE
feature is masked out by default, but then erroneously written back
during VMX initialisation. This would cause memory corruption problems
during boot for XSAVE-capable systems.
Keir Fraser [Mon, 29 Nov 2010 14:40:55 +0000 (14:40 +0000)]
x86: tighten filter on ptwr_do_page_fault()
Even not-so-recent Linux may, due to post-2.6.18 changes to the
process creation code, cause quite a number (depending on environment
and argument size) of faulting accesses to user space originating from
kernel mode. Generally those happen for non-present pages and would
lead to a nested page fault from guest_get_eff_l1e(). They can be
avoided by checking for PFEC_page_present as long as the guest isn't
running on shadow page tables.
Signed-off-by: Jan Beulich <jbeulich@novell.com> Signed-off-by: Keir Fraser <keir@xen.org>
Keir Fraser [Mon, 29 Nov 2010 14:34:32 +0000 (14:34 +0000)]
x86-64: don't crash Xen upon direct pv guest access to GDT/LDT mapping area
handle_gdt_ldt_mapping_fault() is intended to deal with indirect
accesses (i.e. those caused by descriptor loads) to the GDT/LDT
mapping area only. While for 32-bit segment limits indeed prevent the
function being entered for direct accesses (i.e. a #GP fault will be
raised even before the address translation gets done, on 64-bit even
user mode accesses would lead to control reaching the BUG_ON() at the
beginning of that function.
Fortunately the fix is simple: Since the guest kernel runs in ring 3,
any guest direct access will have the "user mode" bit set, whereas
descriptor loads always do the translations to access the actual
descriptors as kernel mode ones.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Further, relax the BUG_ON() in handle_gdt_ldt_mapping_fault() to a
check-and-bail. This avoids any problems in future, if we don't
execute x86_64 guest kernels in ring 3 (e.g., because we use a
lightweight HVM container).
Keir Fraser [Fri, 26 Nov 2010 14:25:30 +0000 (14:25 +0000)]
xenpaging: increase recently used pages from 4MB to 64MB
Increase recently used pages from 4MB to 64MB. Keeping more pages in
memory allows the guest to make more progress if the paging file spans
the entire guest memory.
Keir Fraser [Fri, 26 Nov 2010 14:23:31 +0000 (14:23 +0000)]
xenpaging: handle temporary out-of-memory conditions during page-in
p2m_mem_paging_prep() should return -ENOMEM if a new page could not be
allocated. This can be handled in xenpaging to retry the
page-in. Right now such condition would stall the guest because the
requested page will not come back, xenpaging simply exits. So
xenpaging could very well retry the allocation forever to rescue the
guest.
Keir Fraser [Fri, 26 Nov 2010 14:22:11 +0000 (14:22 +0000)]
xenpaging: optimize p2m_mem_paging_populate
p2m_mem_paging_populate will always put another request in the ring.
To reduce pressure on the ring, place only required requests in the
ring. If the gfn was already processed by another thread, and the
current vcpu does not need to be paused, p2m_mem_paging_resume will do
nothing with the request. And also xenpaging will drop the request if
the vcpu does not need a wakeup.
Keir Fraser [Fri, 26 Nov 2010 14:21:33 +0000 (14:21 +0000)]
xenpaging: when populating a page, check if populating is already in progress
p2m_mem_paging_populate can be called serveral times from different
vcpus. If the page is already in state p2m_ram_paging_in and has a new
valid mfn, invalidating this new mfn will cause trouble later if
p2m_mem_paging_resume will set the new gfn/mfn pair back to state
p2m_ram_rw. Detect this situation and keep p2m state if the page is
in the process of being still paged-out or already paged-in. In fact,
p2m state p2m_ram_paged is the only state where the mfn type can be
invalidated.
Keir Fraser [Fri, 26 Nov 2010 14:20:39 +0000 (14:20 +0000)]
xenpaging: allow negative num_pages and limit num_pages
Simplify paging size argument. If a negative number is specified, it
means the entire guest memory should be paged out. This is useful for
debugging. Also limit num_pages to the guests max_pages.
Keir Fraser [Fri, 26 Nov 2010 14:20:10 +0000 (14:20 +0000)]
xenpaging: notify policy only on resume
If a page is requested more than once, the policy is also notified
more than once about the page-in. However, a page-in happens only
once. Any further resume will only unpause the other vcpu. The
multiple notify will put the page into the mru list multiple times and
it will unlock other already resumed pages too early. In the worst
case, a page that was just resumed can be evicted right away, causing
a deadlock in the guest.
Keir Fraser [Fri, 26 Nov 2010 14:17:56 +0000 (14:17 +0000)]
xenpaging: populate paged-out pages unconditionally in grant code
Populate a page unconditionally to avoid missing a page-in request.
If the page is already in the process of being paged-in, the this vcpu
will be stopped and later resumed once the page content is usable
again.
This matches other p2m_mem_paging_populate usage in the source tree.
projects / xen.git / log